Recently, everyone has been hyping up AI Agents to handle all on-chain operations, but I'm actually more concerned about which parts still require human oversight. For example, authorization/signing—that is, essentially handing over the keys. No matter how smart the Agent is, it can't take responsibility for the consequences; and for cross-chain transfers, modifying contract parameters, shutting down a strategy—one wrong step and there's no turning back. It's better for a human to review everything. There's too much noise in on-chain information, and when KOLs coordinate, it feels more like a radio drama. Thinking about it later, it’s quite ridiculous. By the way, during the recent upgrade of a major public chain, the group was guessing whether projects would migrate. That made me feel even more: tools can be automated, but the questions of “when to stop” and “who to trust or not trust” still rely on humans tightening the reins for now.